Re: [PATCH] coredump: Retry writes where appropriate

[Oleg Nesterov <oleg@redhat.com>]

On 06/02, Alan Cox wrote:
>
> > Perhaps it is easier to change dump_write() to clear TIF_SIGPENDING
> > unless fatal_signal_pending(),
>
> If you are receiving a continuous stream of SIGIO's say then how do you
> guarantee the code below will make progress ?

Yes, the second SIGIO has no effect.

> > 	int coredump_file_write(struct file *file, const void *addr, int nr)
> > 	{
> > 		while (nr > 0) {
> > 			int res = file->f_op->write(file, addr, nr, &file->f_pos);
> >
> > 			if (res > 0) {
> > 				nr -= res;
> > 				continue;
> > 			}
> >
> > 			if (!signal_pending(current))
> > 				break;
> > 			if (__fatal_signal_pending(current))
> > 				break;
> > 			clear_thread_flag(TIF_SIGPENDING);
> > 		}
> >
> > 		return !nr;
> > 	}
> >
>
> > Of course, this all assumes f_op->write() does not do recalc_sigpending().
>
> Which is itself a dangerous assumption that shouldn't be propogated.

I don't think this assumption is dangerous. Why should ->write() call
recalc_sigpending() ? It should not be called outside of signal code.

But yes, if we add SIGNAL_GROUP_DUMPING we can change recalc_sigpending_tsk().
Just I don't like the idea to slow down / complicate this helper for the
very special case.


Hmm. Does ->core_dump() report the state of ->sighand->action? I can't
find any usage. Perhaps we can just ignore all signals except SIGKILL ?

Oleg.